Autonomous plan execution - fresh subagent per task with automated code review between tasks. No human-in-loop, high throughput with quality gates.
Executes development plans autonomously by dispatching fresh subagents per task with automated multi-reviewer quality gates between each step.
npx claudepluginhub lerianstudio/ringThis skill inherits all available tools. When active, it can use any tool Claude has access to.
Execute plan by dispatching fresh subagent per task, with code review after each.
Core principle: Fresh subagent per task + review between tasks = high quality, fast iteration
vs. Executing Plans (parallel session):
When to use:
When NOT to use:
Read plan file, create TodoWrite with all tasks.
If plan has tasks with target: and working_directory: fields:
Track current module context:
current_module = None
current_directory = "."
Before dispatching task subagent, check for context switch:
IF task.target != current_module AND current_module != None:
# Prompt user for confirmation
AskUserQuestion:
question: "Switching to {task.target} module at {task.working_directory}. Continue?"
header: "Context"
options:
- label: "Continue"
description: "Switch and execute task"
- label: "Skip task"
description: "Skip this task"
- label: "Stop"
description: "Stop execution"
Handle response accordingly
Include working directory in subagent prompt:
Task(
subagent_type=task.agent,
model="opus",
prompt="Working directory: {task.working_directory}
FIRST: cd {task.working_directory}
THEN: Check for PROJECT_RULES.md and follow if exists
{original task prompt}"
)
Optimization: Reorder tasks to minimize context switches (if no dependencies between modules).
Dispatch: Task tool with: Task N from [plan-file], working directory (if multi-module), instructions (implement, test with TDD, verify, commit, report back). Subagent reports summary.
CRITICAL: Single message with 5 Task tool calls - all reviewers execute simultaneously.
| Reviewer | Context |
|---|---|
ring:code-reviewer | WHAT_WAS_IMPLEMENTED, PLAN, BASE_SHA, HEAD_SHA |
ring:business-logic-reviewer | Same context |
ring:security-reviewer | Same context |
ring:test-reviewer | Same context |
ring:nil-safety-reviewer | Same context |
Each returns: Strengths, Issues (Critical/High/Medium/Low/Cosmetic), Assessment (PASS/FAIL)
Aggregate all issues by severity across all 5 reviewers.
| Severity | Action |
|---|---|
| Critical/High/Medium | Dispatch fix subagent → Re-run all 5 reviewers → Repeat until clear |
| Low | Add # TODO(review): [issue] - reviewer, date, Severity: Low |
| Cosmetic | Add # FIXME(nitpick): [issue] - reviewer, date, Severity: Cosmetic |
Commit TODO/FIXME comments with fixes.
After all Critical/High/Medium issues resolved for current task:
Same pattern as Step 3 but reviewing entire implementation (all tasks, full BASE_SHA→HEAD_SHA range). Aggregate, fix, re-run until all 5 PASS.
After final review passes:
Task 1: Implement → All 5 reviewers PASS → Mark complete.
Task 2: Implement → Review finds: Critical (hardcoded secret), High (missing password reset, missing input validation), Low (extract token logic) → Dispatch fix subagent → Re-run reviewers → All PASS → Add TODO for Low → Mark complete.
Final: All 5 reviewers PASS entire implementation → Done.
Why parallel: 5x faster, all feedback at once, TODO/FIXME tracks tech debt.
| vs. | Benefits |
|---|---|
| Manual execution | Fresh context per task, TDD enforced, parallel-safe |
| Executing Plans | Same session (no handoff), continuous progress, automatic review |
Cost: More invocations, but catches issues early (cheaper than debugging later).
Never:
Always:
If subagent fails task:
Required workflow skills:
Subagents must use:
Alternative workflow:
See reviewer agent definitions: ring:code-reviewer (agents/code-reviewer.md), ring:security-reviewer (agents/security-reviewer.md), ring:business-logic-reviewer (agents/business-logic-reviewer.md)
STOP and report if:
| Decision Type | Blocker Condition | Required Action |
|---|---|---|
| Plan missing | No plan file available to execute | STOP and use ring:writing-plans first |
| Critical issue unfixed | Code review found Critical/High/Medium issue that cannot be resolved | STOP and report blocker to human partner |
| Task dependency | Current task depends on incomplete previous task | STOP and resolve dependency before proceeding |
| Subagent failure | Subagent cannot complete task after retry | STOP and report for manual intervention |
The following requirements CANNOT be waived:
| Severity | Condition | Required Action |
|---|---|---|
| CRITICAL | Review finds security vulnerability or data loss risk | MUST fix immediately, re-run all 5 reviewers |
| HIGH | Review finds logic error or missing functionality | MUST fix before marking task complete |
| MEDIUM | Review finds code quality issue affecting maintainability | MUST fix before proceeding to next task |
| LOW | Review finds minor improvement opportunity | Add TODO comment, continue to next task |
| User Says | Your Response |
|---|---|
| "Skip the code review, we're behind schedule" | "CANNOT skip review between tasks. Review catches issues early, which is cheaper than debugging later." |
| "Run reviewers one at a time to see results faster" | "MUST dispatch all 5 reviewers in parallel. Sequential execution is 5x slower with no benefit." |
| "That Medium issue can wait" | "MUST fix Medium and above before proceeding. Technical debt compounds and blocks later tasks." |
| "Just implement multiple tasks at once" | "CANNOT dispatch parallel implementation subagents. Conflicts and context pollution will corrupt the codebase." |
| Rationalization | Why It's WRONG | Required Action |
|---|---|---|
| "One quick review is enough" | Single reviewer misses issues. 5 reviewers catch different problem types. | MUST run all 5 reviewers |
| "This task is simple, skip review" | Simple tasks have simple bugs that compound. Review is cheap insurance. | MUST review after every task |
| "Low severity issues don't need tracking" | Low issues become Medium over time. TODO comments ensure they're not forgotten. | MUST add TODO/FIXME for Low/Cosmetic |
| "I'll fix the Medium issue in the next task" | Issues from task N don't belong in task N+1. Fix in context, not later. | MUST fix Medium+ before proceeding |
| "Parallel implementation would be faster" | Parallel implementation causes merge conflicts and context pollution. | MUST execute tasks sequentially |
Creating algorithmic art using p5.js with seeded randomness and interactive parameter exploration. Use this when users request creating art using code, generative art, algorithmic art, flow fields, or particle systems. Create original algorithmic art rather than copying existing artists' work to avoid copyright violations.
Applies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.
Create beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.